RadeonSI Now Uses The HSA ABI For Some Compute Shaders

LLVM/GPGPU expert Tom Stellard has landed a seemingly big change in Mesa for benefiting the RadeonSI driver and step-by-step still advancing the open-source AMD GPU compute support.

On Friday, Stellard landed a change to the RadeonSI compute support for using the HSA (Heterogeneous System Architecture) ABI for non-TGSI (the Gallium3D IR) compute shaders. It switches non-TGSI compute shaders to using the HSA ABI as documented via the Radeo Open Compute ROCm stack.

Tom commented about the HSA ABI, "[it] provides a much cleaner interface for compute shaders and allows us to share more code in the compiler with the HSA stack."

Though at this stage it doesn't look like it will mean much for end-users but any progress on the open-source Radeon compute/OpenCL support or the further maturing of the RadeonOpenCompute code is certainly much welcome as it's one of the few areas where the AMDGPU-PRO/Catalyst stack is still better than the current open-source code. Plus how many years now have we been waiting for open-source desktop developers to better embrace OpenCL/HSA?